Software Design

Deriving a solution which satisfies software Speficitaion

Stages of Design

Problem understanding– Look at the problem from different angles to discover the

design requirements. Identify one or more solutions

– Evaluate possible solutions and choose the most appropriate depending on the designer's experience and available resources.

Describe solution abstractions– Use graphical, formal or other descriptive notations to

describe the components of the design. Repeat process for each identified abstraction

until the design is expressed in primitive terms.

The Design Process

Any design may be modelled as a directed graph made up of entities with attributes which participate in relationships.

The system should be described at several different levels of abstraction.

Design takes place in overlapping stages. It is artificial to separate it into distinct phases but some separation is usually necessary.

Phases in the Design Process

Architecturaldesign

Abstractspecificatio

n

Interfacedesign

Componentdesign

Datastructuredesign

Algorithmdesign

Systemarchitecture

Softwarespecification

Interfacespecification

Componentspecification

Datastructure

specification

Algorithmspecification

Requirementsspecification

Design activities

Design products

Design Phases

• Architectural design: Identify sub-systems.

• Abstract specification: Specify sub-systems.

• Interface design: Describe sub-system interfaces.

• Component design: Decompose sub-systems into components.

• Data structure design: Design data structures to hold problem data.

• Algorithm design: Design algorithms for problem functions.

Design

Computer systems are not monolithic: they are usually composed of multiple, interacting modules.

Modularity has long been seen as a key to cheap, high quality software.

The goal of system design is to decode:– What the modules are;

– What the modules should be;

– How the modules interact with one-another

Modular programming

In the early days, modular programming was taken to mean constructing programs out of small pieces: “subroutines”

But modularity cannot bring benefits unless the modules are – autonomous, – coherent and – robust

Procedural Abstraction

The most obvious design methods involve functional decomposition.

This leads to programs in which procedures represent distinct logical functions in a program.

Examples of such functions:– “Display menu”– “Get user option”

This is called procedural abstraction

Programs as Functions

Another view is programs as functions:

input output

x f f (x) the program is viewed as a function from a set I of legal inputs to a set O of outputs.

There are programming languages (ML, Miranda, LISP) that directly support this view of programming

Well-suited to certain application domains - e.g., compilers

Less well-suited to distributed, non-terminating systems - e.g., process control systems, operating systems like WinNT, ATM machines

Object-oriented design

– The system is viewed as a collection of interacting objects.

– The system state is decentralized and each object manages its own state.

– Objects may be instances of an object class and communicate by exchanging

methods.

Five Criteria for Design Methods

We can identify five criteria to help evaluate modular design methods:– Modular decomposability;– Modular composability;– Modular understandability;– Modular continuity;– Modular protection.

Modular Decomposability

This criterion is met by a design method if the method supports the decomposition of a problem into smaller sub-problems, which can be solved independently.

In general method will be repetitive: sub-problems will be divided still further

Top-down design methods fulfil this criterion; stepwise refinement is an example of such method

Hierarchical Design Structure

System level

Sub-systemlevel

Top-down Design

• In principle, top-down design involves starting

at the uppermost components in the hierarchy and working down the hierarchy level by level.

• In practice, large systems design is never truly top-down. Some branches are designed before others. Designers reuse experience (and sometimes components) during the design process.

Modular Composability

A method satisfies this criterion if it leads to the production of modules that may be freely combined to produce new systems.

Composability is directly related to the issue of reusability Note that composability is often at odds with

decomposability; top-down design, – for example, tends to produce modules that may not be composed

in the way desired This is because top-down design leads to modules which

fulfil a specific function, rather than a general one

Examples

The Numerical Algorithm Group (NAG) libraries contain a wide range of routines for solving problems in linear algebra, differential equations, etc.

The Unix shell provides a facility called a pipe, written “”, whereby – the standard output of one program may be

redirected to the standard input of another; this convention favours composability.

Modular Understandability

A design method satisfies this criterion if it encourages the development of modules which are easily understandable.

COUNTER EXAMPLE 1. Take a thousand lines program, containing no procedures; it’s just a long list of sequential statements. Divide it into twenty blocks, each fifty statements long; make each block a method.

COUNTER EXAMPLE 2. “Go to” statements.

Understandability

• Related to several component characteristics– Can the component be understood on its own?– Are meaningful names used?– Is the design well-documented?– Are complex algorithms used?

• Informally, high complexity means many relationships between different parts of the design.

Modular Continuity

A method satisfies this criterion if it leads to the production of software such that a small change in the problem specification leads to a change in just one (or a small number of ) modules.

EXAMPLE. Some projects enforce the rule that no numerical or textual literal should be used in programs: only symbolic constants should be used

COUNTER EXAMPLE. Static arrays (as opposed to open arrays) make this criterion harder to satisfy.

Modular Protection

A method satisfied this criterion if it yields architectures in which the effect of an abnormal condition at run-time only effects one (or very few) modules

EXAMPLE. Validating input at source prevents errors from propagating throughout the program.

COUNTER EXAMPLE. Using int types where subrange or short types are appropriate.

Five principles for Good Design

From the discussion above, we can distil five principles that should be adhered to:– Linguistic modular units;– Few interfaces;– Small interfaces– Explicit interfaces;– Information hiding.

Linguistic Modular Units

A programming language (or design language) should support the principle of linguistic modular units:– Modules must correspond to linguistic units in the language used

EXAMPLE. Java methods and classes COUNTER EXAMPLE. Subroutines in BASIC are called

by giving a line number where execution is to proceed from; there is no way of telling, just by looking at a section of code, that it is a subroutine.

Few Interfaces

This principle states that the overall number of communication channels between modules should be as small as possible:– Every module should communicate with as few others as

possible.

So, in the system with n modules, there may be a minimum of n-1 and a maximum of links; your system should stay closer to the minimum

2

)1( nn

Few Interfaces

Small Interfaces (Loose Coupling)

This principle states:– If any two modules communicate, they should exchange as

little information as possible.

COUNTER EXAMPLE. Declaring all instance variables as public!

A measure of the strength of the inter-connections between system components.

Loose coupling means component changes are unlikely to affect other components.– Shared variables or control information exchange lead

to tight coupling.– Loose coupling can be achieved by state

decentralization (as in objects) and component communication via parameters or message passing.

Coupling

Tight Coupling

Module A Module B

Module C Module D

Shared dataarea

Loose Coupling

Module A

A’s data

Module B

B’s data

Module D

D’s data

Module C

C’s data

• Object-oriented systems are loosely coupled because there is no shared state and objects communicate using message passing.

• However, an object class is coupled to its super-classes. Changes made to the attributes or operations in a super-class propagate to all sub-classes.

Coupling and Inheritance

Reusability

A major obstacle to the production of cheap quality software is the intractability of the reusability issue.

Why isn’t writing software more like producing hardware? Why do we start from scratch every time, coding similar problems time after time after time?

Obstacles:– Economic;

– Organizational;

– Psychological.

Stepwise Refinement

The simplest realistic design method, widely used in practice.

Not appropriate for large-scale, distributed systems: mainly applicable to the design of methods.

Basic idea is:– Start with a high-level spec of what a method is to achieve;– Break this down into a small number of problems (usually no

more than 10)– For each of these problems do the same;– Repeat until the sub-problems may be solved immediately.

Explicit Interfaces

If two modules must communicate, they must do it so that we can see it:– If modules A and B communicate, this must be

obvious from the text of A or B or both.

Why? If we change a module, we need to see what other modules may be affected by these changes.

Information Hiding

This principle states:– All information about a module, (and particularly how the

module does what it does) should be private to the module unless it is specifically declared otherwise.

Thus each module should have some interface, which is how the world sees it anything beyond that interface should be hidden.

The default Java rule:– Make everything private

Cohesion

A measure of how well a component “fits together”.

A component should implement a single logical entity or function.

Cohesion is a desirable design component attribute as when a change has to be made, it is localized in a single cohesive component.

Various levels of cohesion have been identified.

Cohesion Levels

Coincidental cohesion (weak)– Parts of a component are simply bundled together.

Logical association (weak)– Components which perform similar functions are grouped.

Temporal cohesion (weak)– Components which are activated at the same time are

grouped.

Cohesion Levels

Communicational cohesion (medium)– All the elements of a component operate on the same input or

produce the same output. Sequential cohesion (medium)

– The output for one part of a component is the input to another part. Functional cohesion (strong)

– Each part of a component is necessary for the execution of a single function.

Object cohesion (strong)– Each operation provides functionality which allows object

attributes to be modified or inspected.

Cohesion as a Design Attribute

Not well-defined. Often difficult to classify cohesion.

Inheriting attributes from super-classes weakens cohesion.– To understand a component, the super-classes

as well as the component class must be examined.

– Object class browsers assist with this process.

Natural Language

Nouns suggest candidate Classes. Not every noun is an Object Class.

– Some are attributes of another Class.

– Some are irrelevant, outside the scope of the application.

Verb phrases suggest class associations– some relationships are irrelevant (caution).

Proper nouns suggest Objects of a Class type. Beware of singular nouns.

Class Description

Develop a Class description, either in textual prose or some other structured form. E.G. using a customer in a Bank– Customer: a holder of one or more accounts in a Bank. A

customer can consist of one or more persons or companies. A customer can: make withdrawals; deposit money; transfer money between their accounts or to another account; query their accounts.

Structured Class Description

Class Name: CustomerDescription: Personal or company detailsSuperclass: User

Name: NameDescription: Customer’s nameType: String (max. 12 chars)Cardinality: 1

Name: OwnsDescription: Details of bank accountsType: AccountCardinality: Many

Structured Class Description (cont..)

Public Methods:

Name: Pay_billParameters: amount, date, destination, account.Description: Customer may pay bills through the Bank.

Structured Class Description (cont..)

Private Methods:

Name: TransferParameters: amount, from_account, to_account.Description: Allow transfers from owned accounts to any others.